PDF OCR & Text Extractor: Text from PDFs, Scans and Images

Turn PDFs, scanned documents and photos of documents into searchable text, one row per page. Pages with a text layer are read directly; scanned pages and images go through Tesseract OCR.

Common uses

• PDF to text in bulk for search, RAG and LLM pipelines: one row per page, with the page number.
• Scanned PDF and image OCR (TIFF, PNG, JPEG): turn scans into searchable text.
• A cheap OCR API: $4 per 1,000 OCR pages, $0.80 per 1,000 text-layer pages, and failed pages are free.

Example inputs

Paste any of these into the JSON input tab:

Text-layer PDF (IRS W-9, 6 pages, about $0.005)

{"urls": ["https://www.irs.gov/pub/irs-pdf/fw9.pdf"], "mode": "text"}

Scanned image OCR (Tesseract test images, about $0.008)

{"urls": ["https://raw.githubusercontent.com/tesseract-ocr/test/main/testing/phototest.tif", "https://raw.githubusercontent.com/tesseract-ocr/tessdoc/main/images/eurotext.png"], "mode": "auto"}

PDF with tables (IRS W-4, 5 pages)

{"urls": ["https://www.irs.gov/pub/irs-pdf/fw4.pdf"], "mode": "auto", "extractTables": true}

Input

• File URLs: public links to PDFs or images (PNG, JPEG, TIFF, GIF, WebP, BMP). The type is detected from the file contents, not the extension.
• Files from a key-value store (optional): pick one of your stores; each record is processed (filter with Key prefix). The Actor can read only the store you pick.
• URLs from a dataset (optional): the url field of each item.
• Mode: auto (text layer, OCR only where it is missing), text (never OCR), ocr (OCR every page).
• Also extract tables as CSV (off by default, free, best effort).

Output

One page row per extracted page: file, page, method (text or ocr), text, charCount, ocrMeanConfidence (OCR only), and tables (only when table extraction is on). Each file also gets a free file-summary row (pagesExtracted, pagesFailed, pagesBilled, truncatedAtPageCap). Problems come back as free page-error / file-error rows with an errorCode.

Pricing (pay per event)

Event	Price
Text page (PDF text layer)	$0.80 per 1,000 pages
OCR page or image (images up to 10 MB)	$4 per 1,000
Large image OCR (10–20 MB)	$8 per 1,000

You pay only for pages actually extracted, never for the page count a file claims. Failed pages and files are free. Free-plan accounts get up to 100 pages per run. Your maximum charge per run is respected: the run stops when it is reached and nothing past it is charged.

Limits

• PDFs up to 50 MB and 500 pages per file (later pages are skipped and reported in the summary).
• Images up to 20 MB and 50 megapixels; large images are scaled to about 3,500 px on the long edge before OCR.
• Each file is processed in its own sandboxed process with a time limit; damaged, password-protected or malicious files return a free error row.
• English OCR. OCR text is a searchable draft, not exact field values: check numbers from receipts or invoices.

Privacy

Files are processed in your own run and not stored or viewed by us. You're responsible for having the right to process them. The Actor never logs extracted text, only counts and error codes.

Open-source components

pypdf (BSD-3), pdfplumber (MIT), Pillow (MIT-CMU), Tesseract (Apache-2.0), poppler pdftoppm (GPL, run as a separate program).